Crop Productivity Analysis

Data and Assumptions

Data

The following datasets are utilized in this analysis for calculating and mapping crop productivity over the past years:

1. Dynamic World Dataset:

   • Source: Dynamic World - Google and the World Resources Institute (WRI)

   • Description: The Dynamic World dataset provides a near real-time, high-resolution (10-meter) global land cover classification. It is derived from Sentinel-2 imagery and utilizes machine learning models to classify land cover into nine distinct classes, including water, trees, grass, crops, built areas, bare ground, shrubs, flooded vegetation, and snow/ice. The dataset offers data with minimal latency, enabling near-immediate analysis and decision-making.

   • Spatial Resolution: 10 meters.

   • Temporal Coverage: Data is available since mid-2015, updated continuously as Sentinel-2 imagery becomes available capturing near Real-time.

2. MODIS Dataset:

   • Source: NASA’s Moderate Resolution Imaging Spectroradiometer MODIS on Terra and Aqua satellites.

   • Description: The MODIS dataset provides a wide range of data products, including land surface temperature, vegetation indices, and land cover classifications. It is widely used for monitoring and modeling land surface processes.

   • Spatial Resolution: 250 meters.

   • Temporal Coverage: Data is available from 2000 to the present, with daily to 16-day composite products.

3. GDP and Agricultural Data:

   • Source: Official government statistics and international databases.

   • Description: This dataset includes regional GDP figures, agricultural output, export data, and crop production statistics. It is used to analyze the economic impact of agricultural productivity.

   • Spatial Resolution: Varies by dataset; typically available at national and subnational levels.

   • Temporal Coverage: Varies by dataset; typically available annually or quarterly.

4. Administrative Boundaries:

   • Source: MIMU (Myanmar Information Management Unit)

   • Description: This dataset provides the administrative boundaries for Myanmar at various levels (national, state/region, district, township). It is used for spatial aggregation and analysis of crop productivity and economic data.

   • Spatial Resolution: Varies by administrative level.

Assumptions and Methodological Notes

This analysis is based on the following assumptions and methodological considerations:

Temporal Definitions and Aggregations

1. Fiscal Year Definition: The first quarter (Q1) of each fiscal year starts in April and ends in March of the following year. This aligns with Myanmar’s fiscal calendar.

2. Quarterly Aggregations: Monthly data are aggregated into quarters (Q1: April-June, Q2: July-September, Q3: October-December, Q4: January-March).

Methodological Assumptions

1. Spatial Aggregation: Regional GDP values are based on single-year reference data and are assumed to maintain consistent spatial distributions across all years in the analysis period.

2. EVI Processing:

   • EVI values are median-aggregated within administrative boundaries to reduce noise

   • Time series preprocessing follows TIMESAT methodology: outlier removal, interpolation, and Savitzky-Golay smoothing

   • Seasonality parameters (start of season, middle of season, end of season) are extracted using a 20% threshold of maximum EVI

3. Cropland Identification: Pixels are classified as cropland based on Dynamic World’s machine learning classification. Classification accuracy varies by region and land cover type, with potential confusion between cropland and similar land covers.

Cropland Area Analysis

Cropland Statistics by Administrative Level 1

The table below shows the cropland area statistics by Administrative Level 1 for the years 2020 and 2025, along with the percentage change and absolute percentage change between these years. Except for Chin State, all regions experienced a decrease in cropland area from 2020 to 2025.

	Name	PCODE	Crop Area (ha) in 2020	Crop Area (ha) in 2025	% Change in Crop Area (2020-2025)	Absolute % Change in Crop Area (2020-2025)
0	Sagaing	MMR005	1,523,048	1,355,659	-10.99%	10.99%
1	Ayeyarwady	MMR017	1,284,046	1,185,667	-7.66%	7.66%
2	Mandalay	MMR010	994,493	919,394	-7.55%	7.55%
3	Magway	MMR009	911,786	832,402	-8.71%	8.71%
4	Bago (East)	MMR007	675,028	612,430	-9.27%	9.27%
5	Yangon	MMR013	465,210	469,040	0.82%	0.82%
6	Bago (West)	MMR008	463,752	442,237	-4.64%	4.64%
7	Shan (South)	MMR014	365,328	300,314	-17.80%	17.80%
8	Mon	MMR011	168,289	178,099	5.83%	5.83%
9	Kachin	MMR001	214,885	164,590	-23.41%	23.41%
10	Shan (North)	MMR015	216,955	154,755	-28.67%	28.67%
11	Rakhine	MMR012	147,233	142,034	-3.53%	3.53%
12	Nay Pyi Taw	MMR018	136,528	117,129	-14.21%	14.21%
13	Kayin	MMR003	104,489	82,406	-21.13%	21.13%
14	Shan (East)	MMR016	40,675	36,171	-11.07%	11.07%
15	Tanintharyi	MMR006	19,499	12,136	-37.76%	37.76%
16	Kayah	MMR002	36,523	10,243	-71.96%	71.96%
17	Chin	MMR004	8,095	10,230	26.38%	26.38%

The following chart illustrates the changes in cropland area across different Administrative Level 1 regions from 2016 to 2025.

Satellite-Derived vs. Official Harvested Area Comparison

The table below compares the satellite-derived cropland area estimates with official harvested area statistics in 2023. Compared to the official harvested area, the cropland area derived from satellite data is significantly lower, with the highest difference observed in Tanintharyi region.

	Admin Level 1	year	Actual Harvested Area (acres)	Satellite-derived Crop Area (acres)	Actual Rank	Satellite-derived Rank	Percent Difference (%)
30	Sagaing	2023	15,737,204	3,962,762	1	1	-297.13%
31	Ayeyarwady	2023	15,600,272	3,029,523	2	2	-414.94%
32	Bago	2023	13,694,134	2,708,850	3	3	-405.53%
33	Mandalay	2023	9,080,864	2,645,312	5	4	-243.28%
34	Magway	2023	9,900,750	2,461,605	4	5	-302.21%
35	Shan	2023	7,783,626	1,762,997	6	6	-341.50%
36	Yangon	2023	4,259,410	1,120,116	7	7	-280.26%
37	Kachin	2023	1,905,864	588,800	11	8	-223.69%
38	Mon	2023	2,896,838	371,377	8	9	-680.03%
39	Nay Pyi Taw	2023	937,660	346,262	13	10	-170.80%
40	Rakhine	2023	2,647,330	320,716	10	11	-725.44%
41	Kayin	2023	2,737,796	188,712	9	12	-1350.78%
42	Kayah	2023	445,018	65,761	14	13	-576.72%
43	Tanintharyi	2023	1,608,060	36,513	12	14	-4304.06%
44	Chin	2023	272,288	24,694	15	15	-1002.64%

In terms of ranking by cropland area, both official harvested area and satellite-derived estimates show similar trends across regions, especially for Sagaing, Ayeyarwady, and Bago regions. However, there are several mismatches in region with smaller cropland area.

Crop Seasonality

Using this time series dataset of EVI images, we apply several pre-processing steps to extract critical phenological parameters: start of season (SOS), middle of season (MOS), end of season (EOS), length of season (LOS), etc. This workflow is heavily inspired by the TIMESAT software.

Pre-processing steps

1. Remove outliers from dataset on per-pixel basis using median method: outlier if median from a moving window < or > standard deviation of time-series times 2.

2. Interpolate missing values linearly

3. Smooth data on per-pixel basis (using Savitsky Golay filter, window length of 3, and polyorder of 1)

Phenology Process
We then extract crop seasonality metrics using the seasonal amplitude method from the phenolopy package.

The chart below shows the result of this process for a single crop pixel. The blue dots represent the raw EVI values, the black line represents the processed EVI values, and the dotted lines represent season parameters extracted for that pixel: start of season, peak of season, and end of season.

Based on the phenology process, we identified the seasonality to start in July and end in February with the peak being in October. This can vary with geographic region and crop type as well, however, that has not been taken into consideration in this version.

EVI Time Series Analysis

National Level Analysis

The following figure shows the median EVI in Myanmar from 2010 to 2025 during the crop growing season (June to February). In general, the median EVI increases from 2010 to 2025.

State/Region Level Analysis (Admin 1)

The figure below shows the trends of median EVI from 2010-2025 on admin level 1.

The figure below shows a choropleth maps of EVI for each admin level 1 from 2010-2025

The figure below shows the percent change of EVI compared to previous year from 2010-2025 on admin level 1.

District Level Analysis (Admin 2)

The figure below shows a choropleth maps of EVI for each admin level 2 from 2010-2025

EVI and Economic Indicators

EVI and Overall GDP

The figure below shows the relationship between national median EVI and overall GDP from 2010 to 2025, with a weak positive correlation.

EVI and Agricultural GDP

The figure below shows quarterly EVI (lagged one quarter) and agricultural GDP from 2010 to early 2025. The near-identical cyclical patterns suggest agricultural GDP materializes approximately one quarter later.

The figure below shows annual agricultural GDP and EVI from 2010 to early 2025. While the EVI remains steady, agricultural GDP shows a declining trend since 2020, likely due to external factors such as political instability.

The figure below shows the relationship between agricultural GDP and EVI with various lags. There’s a strong positive correlation between EVI one-quarter lag and agricultural GDP.

The table below presents regression results for national median EVI and agricultural GDP. Lagged EVI (one quarter) is statistically significant across all specifications. In the full model, a 1% increase in EVI corresponds to a 1.12% increase in agricultural GDP, holding other variables constant.

	Dependent variable: gdp_log
	Current EVI only	Current + 1 Quarter Ago	Current + 2 Quarters Ago	Current + 1 Quarter Ago + Is Crop Season
	(1)	(2)	(3)	(4)
EVI (log)	1.701***	1.173***	-0.009	-0.396***
	(0.447)	(0.311)	(0.547)	(0.147)
NTL Sum (log)		-0.937***	-0.642***	-0.540***
		(0.220)	(0.238)	(0.086)
EVI 1 Quarter Ago (log)		2.334***	2.540***	1.115***
		(0.299)	(0.294)	(0.132)
EVI 2 Quarter Ago (log)			-1.557**
			(0.607)
Is Crop Season				1.484***
				(0.086)
Observations	60	53	53	53
R2	0.200	0.705	0.741	0.959
Adjusted R2	0.186	0.687	0.719	0.956
Residual Std. Error	0.749 (df=58)	0.466 (df=49)	0.441 (df=48)	0.175 (df=48)
F Statistic	14.463*** (df=1; 58)	39.098*** (df=3; 49)	34.311*** (df=4; 48)	283.827*** (df=4; 48)
Note:	*p<0.1; **p<0.05; ***p<0.01

State-level EVI and Agricultural GDP

The table below summarizes cropland statistics aggregated at Administrative Level 1 (State/Region) for the year 2015-2016. The regional GDP values across all years in the analysis period are based on this single-year reference data and are assumed to maintain consistent spatial distributions.

Regional GDP values are based on single-year reference data and are assumed to maintain consistent spatial distributions across all years in the analysis period.

	adm1_name	Agriculture
0	Ayeyarwady	16.0%
1	Chin	0.5%
2	Kachin	2.1%
3	Kayah	0.5%
4	Kayin	2.4%
5	Magway	14.1%
6	Mandalay	8.3%
7	Mon	4.4%
8	Nay Pyi Taw	1.1%
9	Rakhine	3.3%
10	Sagaing	17.0%
11	Tanintharyi	6.9%
12	Yangon	4.9%
13	Bago (East)	5.3%
14	Bago (West)	5.3%
15	Shan (South)	2.7%
16	Shan (East)	2.7%
17	Shan (North)	2.7%

The figure below shows the relationship between state-level median EVI and agricultural GDP across several lags. The positive correlation between lagged EVI (one quarter) and agricultural GDP persists at the state level.

The chart shows the relationship between state-level median EVI and agricultural GDP across states/regions. Colors differentiate data for the crop season versus the non-crop season. The figure suggests that during crop season, all states shows a positive relationship between EVI and agricultural GDP.

The table presents regression results for state-level median EVI and agricultural GDP.

	Dependent variable: gdp_log
	Current EVI only	Current + NTL	Current + NTL + 1 Quarter Ago	Current + NTL + 1 Quarter Ago + Is Crop Season	All Variables + Region Fixed Effects
	(1)	(2)	(3)	(4)	(5)
Intercept	11.544***	11.028***	11.503***	6.824***	13.923***
	(0.176)	(0.466)	(0.466)	(0.405)	(0.174)
EVI (log)	0.096	0.125	0.027	-1.055***	-0.244***
	(0.137)	(0.149)	(0.148)	(0.122)	(0.036)
EVI 1 Quarter Ago (log)			0.837***	-0.099	0.674***
			(0.148)	(0.120)	(0.035)
Is Crop Season				2.119***	1.602***
				(0.084)	(0.024)
NTL Sum (log)		0.053	0.097**	0.150***	-0.175***
		(0.045)	(0.045)	(0.035)	(0.016)
Observations	1057	938	937	937	937
R2	0.000	0.002	0.035	0.425	0.963
Adjusted R2	-0.000	-0.000	0.032	0.423	0.962
Residual Std. Error	1.264 (df=1055)	1.270 (df=935)	1.250 (df=933)	0.965 (df=932)	0.248 (df=915)
F Statistic	0.495 (df=1; 1055)	0.871 (df=2; 935)	11.275*** (df=3; 933)	172.378*** (df=4; 932)	1129.960*** (df=21; 915)
Note:	*p<0.1; **p<0.05; ***p<0.01

EVI and Crop Exports

The figure below shows the relationship between national median EVI and total exports by crop. Except for beans and pulses, all crops exhibit positive correlation with lagged EVI (one quarter). Some crops show stronger relationships at a two-quarter lag, suggesting crop-specific export materialization timings.

The table below shows the regression result for total vegetable exports with various model specifications. Lagged EVI (one quarter) is statistically significant in all models. In the full model, a 1% increase in EVI corresponds to a 0.45% increase in total vegetable exports, holding other variables constant.

	Dependent variable: total_vegetable_exports_log
	Current EVI only	Current + 1 Lag	Current + 2 Lags	All Variables
	(1)	(2)	(3)	(4)
EVI (log)	-0.448***	-0.480***	0.028	0.064
	(0.114)	(0.109)	(0.189)	(0.188)
EVI 1 Quarter Ago (log)		0.348***	0.323***	0.449***
		(0.110)	(0.104)	(0.120)
EVI 2 Quarters Ago (log)			0.658***	0.408*
			(0.192)	(0.227)
Is Crop Season				-0.208**
				(0.103)
Observations	168	165	162	162
R2	0.085	0.152	0.229	0.249
Adjusted R2	0.079	0.141	0.215	0.230
Residual Std. Error	0.389 (df=166)	0.369 (df=162)	0.346 (df=158)	0.343 (df=157)
F Statistic	15.376*** (df=1; 166)	14.510*** (df=2; 162)	15.669*** (df=3; 158)	13.011*** (df=4; 157)
Note:	*p<0.1; **p<0.05; ***p<0.01

The table below shows the regression results of the relationship total vegetables, beans and pulses, corn, and rice exports with national median EVI. Every crop except beans and pulses shows a statistically significant positive relationship with EVI from 1 quarter ago. The beans and pulses export have a significant relationship with EVI from 2 quarters ago.

	Total Vegetables	Beans/Pulses	Corn	Rice
	(1)	(2)	(3)	(4)
Intercept	7.523***	5.800***	5.012***	6.662***
	(0.499)	(0.585)	(1.469)	(0.770)
EVI (log)	0.064	-0.107	-0.249	0.324
	(0.188)	(0.220)	(0.553)	(0.290)
EVI 1 Quarter Ago (log)	0.449***	-0.226	0.679*	0.862***
	(0.120)	(0.141)	(0.354)	(0.185)
EVI 2 Quarters Ago (log)	0.408*	0.777***	0.443	0.232
	(0.227)	(0.266)	(0.668)	(0.350)
Is Crop Season	-0.208**	-0.080	-0.201	-0.170
	(0.103)	(0.120)	(0.302)	(0.158)
Observations	162	162	162	162
R2	0.249	0.310	0.079	0.147
Adjusted R2	0.230	0.292	0.055	0.125
Residual Std. Error	0.343 (df=157)	0.402 (df=157)	1.010 (df=157)	0.529 (df=157)
F Statistic	13.011*** (df=4; 157)	17.618*** (df=4; 157)	3.351** (df=4; 157)	6.751*** (df=4; 157)
Note:	*p<0.1; **p<0.05; ***p<0.01